Skip to content
Lenny's PodcastLenny's Podcast

Keith Coleman & Jay Baxter: How bridging finds neutral truth

Through bridging-based scoring that rewards agreement between users who disagree; only 7% of proposed notes ever ship, and Meta now copies the algorithm.

Lenny RachitskyhostKeith ColemanguestJay Baxterguest
Feb 26, 20251h 47mWatch on YouTube ↗

At a glance

WHAT IT’S REALLY ABOUT

Inside X’s Community Notes: Crowdsourcing Neutral Truth At Internet Scale

  1. The episode explores how X’s Community Notes crowdsources context on potentially misleading posts and uses a novel "bridging" algorithm to surface notes agreed upon by people who usually disagree.
  2. Keith Coleman and Jay Baxter walk through the product’s origin, design principles, and the small, highly autonomous “thermal” team structure that allowed it to survive multiple leadership changes and become an industry model.
  3. They explain why transparency, open participation, and strict quality thresholds are essential for trust, how external research validates its impact on reducing misinformation spread, and how Meta and independent researchers are now adopting and extending the system.
  4. The conversation also highlights broader lessons about lean teams, low‑ego leadership, and why Community Notes reveals there is far more cross‑partisan agreement on facts than the public narrative suggests.

IDEAS WORTH REMEMBERING

5 ideas

Crowdsourced context can rival professional fact-checkers when carefully structured.

Community Notes doesn’t rely on majority votes or experts; instead it surfaces notes rated helpful by people who historically disagree, which produces neutral, verifiable, and widely accepted context.

Quality and trust depend on strict thresholds and visible constraints on power.

Only ~7–10% of proposed notes ever show, and no one at X has a "force override" button; if a bad note shows, they treat it as a system-design failure, not something to hand-fix, which reinforces user trust.

Radical transparency and open participation are core to legitimacy.

The scoring code and full rating data are open-sourced so outsiders can reproduce results, audit for bias, and even propose better algorithms—turning Community Notes into a genuinely community-built system.

Pseudonymity increases honest cross-partisan agreement.

Testing showed contributors were more willing to endorse notes that challenge their own “side” when not tied to their main identity, and harassment risk dropped—contrary to the usual assumption that real names improve discourse.

Lean, fully focused teams move faster and build more impactful products.

A tiny cross-functional "thermal" team (one backend, one frontend, one ML, one designer, one researcher, one PM) with a single senior decision-maker (Elon) iterated rapidly without OKRs, Jira, or heavy process and shipped what an org of hundreds might not have.

WORDS WORTH SAVING

5 quotes

We actually look for agreement from people who have disagreed in the past. That surprising agreement is what makes the notes so neutral and accurate.

Jay Baxter

This thing is going to be the voice of the people. It’s not going to represent the company’s voice.

Keith Coleman

If there’s a problem with a note that’s so bad you want to do something about it, it’s a problem with the system.

Keith Coleman

People a few years ago were pretty pessimistic that fact-checking ever changes people’s understandings. External studies now show Community Notes does.

Jay Baxter

Society often feels really polarized, but Community Notes shows people really can agree on quite a lot—even on super controversial topics.

Keith Coleman

What Community Notes is and how the bridging-based algorithm worksDesign principles: openness, neutrality, anonymity, and voice of the peopleTeam structure and culture (thermal model, lean staffing, high autonomy)Evolution of Community Notes through multiple CEOs and the X acquisitionMeasured impact on misinformation spread, user beliefs, and behaviorUse of open source, external research, and emerging AI/LLM integrationsBroader implications for trust, polarization, and future governance models

High quality AI-generated summary created from speaker-labeled transcript.

Get more out of YouTube videos.

High quality summaries for YouTube videos. Accurate transcripts to search & find moments. Powered by ChatGPT & Claude AI.

Add to Chrome